vt-d: Do dpci eoi outside of irq_lock.

Deadlock may occur if do hvm_dpci_eoi() inside of irq_lock on MP
platform. For example, there are two physical cpus. If interrupt is
injected on cpu0, but vcpu is migrated to cpu1 and it does eoi inside
of irq_lock, then IPI will be issued to cpu0. At the same time, cpu0
may have disabled irq and is acquiring the same irq_lock. In addition,
current code cannot guarantee do hvm_dpci_eoi() inside of irq_lock
when timeout. This patch does hvm_dpci_eoi() outside of irq_lock, and
solves above problems.

Signed-off-by: Xiaohui Xin <xiaohui.xin@intel.com>
Signed-off-by: Weidong Han <weidong.han@intel.com>
Signed-off-by: Keir Fraser <keir@xensource.com>

1 files changed, 7 insertions, 7 deletions

diff --git a/xen/arch/x86/hvm/vpic.c b/xen/arch/x86/hvm/vpic.c
index 161129a97f..4a3fc5ae5a 100644
--- a/xen/arch/x86/hvm/vpic.c
+++ b/xen/arch/x86/hvm/vpic.c
@@ -249,13 +249,13 @@ static void vpic_ioport_write(
                 vpic->isr &= ~(1 << irq);
                 if ( cmd == 7 )
                     vpic->priority_add = (irq + 1) & 7;
-                if ( vtd_enabled )
-                {
-                    irq |= ((addr & 0xa0) == 0xa0) ? 8 : 0;
-                    hvm_dpci_eoi(current->domain,
-                                 hvm_isa_irq_to_gsi(irq), NULL);
-                }
-                break;
+                /* Release lock and EOI the physical interrupt (if any). */
+                vpic_update_int_output(vpic);
+                vpic_unlock(vpic);
+                hvm_dpci_eoi(current->domain,
+                             hvm_isa_irq_to_gsi((addr >> 7) ? (irq|8) : irq),
+                             NULL);
+                return; /* bail immediately */
             case 6: /* Set Priority                */
                 vpic->priority_add = (val + 1) & 7;
                 break;